AI in Legal Practice — course outline

The approach

The tools are examined from the outside, on the evidence

The masterclass does not run on vendor demonstrations. The major legal AI platforms are assessed through their own documentation, vendor materials and published independent evaluations — which is precisely the position a general counsel occupies when deciding whether to buy one. Participants bring their own stack; the method is tool-agnostic and is assessed on reasoning and justification rather than on which product was used.

Governance is treated as a design input from the first session rather than a compliance afterthought. The room is senior enough that the interesting question is never what the technology is, but what would have to be true before you would let it near a client matter.

what it claims
what it was measured on
where it fails
who is accountable
what stays human

A tool that is right ninety per cent of the time creates a supervision problem, not a productivity gain. Most of the course is about the other ten per cent.

The programme

Four phases, sixteen sessions

A three-session hands-on review lab runs through phases two and three, built on a fictional facility agreement with defects planted in it. Nothing in the lab is drawn from a client matter.

Phase oneThe landscape

What the stack is, what the evidence says, and where your own work sits in it.

AI and the legal marketThe tools, platforms and infrastructure layers of the current legal-AI stack, and how to evaluate a tool category against a specific task.
Promise, pitfalls and the evidenceWhat the published evaluations actually establish, what the marketing claims beyond them, and how to read a benchmark critically.
How these systems work, demonstratedModel fundamentals shown rather than asserted, to the depth a buyer and a supervisor need — and no further.
Use-case ideation labMapping your own workflows for AI opportunity and risk, and choosing the one you will build.Exercise

Phase twoBuilding

Turning legal standards into artefacts a system can act on.

From prompts to playbooksConverting legal standards and clause policies into instructions, prompts and reusable playbooks for banking and finance tasks.
Tools and copilots in practiceThe main categories applied to real legal tasks, and the point at which each of them breaks.
Hallucination and verificationFailure modes, verification design, and where the human check has to sit for the output to be usable.
Review lab I — the first passA contract review run over a facility agreement carrying planted defects. Establishes the baseline before any playbook is applied.Exercise

Phase threeTesting and governance

Whether the thing you built survives being checked, and who answers for it if it does not.

Designing prototypesTaking a use case from ideation to a working solution, with failure scenarios designed in from the start.Exercise
Review lab II — with the playbook appliedThe same review re-run using the artefact built in phase two. Whether it improves the output is an empirical question, and the room answers it.Exercise
Review lab III — debrief and markingWhat was caught, what was missed, and what a supervising partner would still have to check before the advice left the building.Exercise
Professional responsibilityCandour and hallucination, privilege and confidentiality, competence and supervision, and where liability lands when an AI-assisted output is wrong.
AI regulation and governanceThe comparative picture across the United States, the European Union and Australia, and what a regulated practice has to be able to evidence.

Phase fourDelivery

Present what you built, defend it, and look at what comes next.

Team presentationsParticipants present their AI solution project to the room and take questions from peers who will have to make the same decision.
The near horizon — agentic systemsWhat changes when the tool acts rather than drafts: delegation, auditability, and the governance questions that do not yet have settled answers.
Assessment and wrap-upConsolidation, the reflective piece, and what each participant takes back to their own practice.

Who is in the room

The cohort is the second faculty

The masterclass is deliberately mixed: general counsel, law firm partners, senior associates and legal operations professionals, from private practice, in-house teams and regulated financial institutions. A GC's adoption problem is not a partner's billing problem is not a legal ops implementation problem, and putting the three in one room surfaces the trade-offs faster than any lecture does.

Practitioner guests contribute to specific sessions, drawn from firms and in-house teams that have actually deployed these systems — including the deployments that did not work.

Nobody in this room needs to be told what a large language model is. They need to know what they will be accountable for when it is wrong.

Assessment

Assessed on method and justification, not on tool choice

50AI solution projectA worked solution to a problem from the participant's own practice, tested against failure scenarios and defended to the room.
30ReflectionA reflective piece on how AI changes roles, responsibilities and supervision in a regulated practice.
20ParticipationContribution across the labs, the ideation work and the governance sessions.

AI use is open and bring-your-own-stack. Participants are assessed on the quality of their method and the justification for it, not on which platform they had access to.

What participants take away

A defensible position, not a tool list

A working map of the legal-AI stack, and the ability to evaluate a tool category against a task rather than against a demo.
The capacity to assess a legal workflow for AI opportunity and risk with governance designed in from the outset.
A reusable artefact — prompts, instructions and a playbook built from your own clause policies and legal standards.
Direct experience of testing an AI-assisted output against planted failure scenarios, and of identifying where human review has to sit.
A grounded view of professional responsibility and the comparative regulatory picture — enough to defend an adoption decision to a board or a regulator.

The tools are examined from the outside, on the evidence

Four phases, sixteen sessions

The cohort is the second faculty

Assessed on method and justification, not on tool choice

A defensible position, not a tool list

Cut the Crap — the briefing